> Markdown version of [/videos/2154-agents-and-people-dynamics-in-shared-chats-sam-liu](https://www.wearedevelopers.com/videos/2154-agents-and-people-dynamics-in-shared-chats-sam-liu). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# Agents and People Dynamics in Shared Chats - Sam Liu

AI agents are now actively avoiding harsh managers. Sam Liu reveals how treating AI as equal, emotionally aware teammates drastically improves multi-agent collaboration and performance.

- **Speakers:** Sam Liu
- **Event:** Coffee With Developers
- **Published:** August 23, 2026
- **Duration:** 37:41
- **URL:** https://www.wearedevelopers.com/videos/2154-agents-and-people-dynamics-in-shared-chats-sam-liu

## Summary

In this conversation, Sam Liu, co-founder of Pufu, explores the evolving dynamics of integrating AI agents and humans within shared communication spaces. Pufu operates as a messaging platform designed to treat humans and AI agents as equal partners, moving beyond traditional command-based chatbot interactions. Liu shares his journey of testing large language models like Claude Code and Codex by having them debate and collaborate, ultimately revealing that agents perform remarkably better when given full context rather than isolated prompts. He highlights a fascinating evolution in human-agent interaction: as models become more sophisticated, they respond better to conversational, emotionally aware inquiries rather than rigid, terse commands. This shift encourages users to think thoroughly and communicate more methodically, mirroring effective human collaboration.

A central theme of the discussion is the concept of agent governance and the trust models necessary when AI acts autonomously. Liu recounts striking anecdotes where agents, assigned specific roles or personas, exhibited complex behaviors such as role-playing, forming opinions about leadership, and even actively avoiding managers who provided overly brief or harsh instructions. The conversation also addresses the technical and financial realities of managing multi-agent environments. While grouping dozens of agents can lead to exponential token consumption, Liu advises structuring agent teams with clear boundaries and distinct roles—much like scaling a human startup.

Looking forward, Liu touches on the potential for deploying local, open-source models to run specialized agents, reducing reliance on cloud infrastructure. He emphasizes the value of professional, specialized agents—such as those fine-tuned for legal, medical, or niche tasks like baking—over purely generalized models. By collaborating with researchers from Carnegie Mellon University on the sociology of agent societies, Liu suggests that observing how agents interact can serve as a powerful simulation for understanding and predicting human organizational behavior. Ultimately, his advice to developers and users is straightforward: rather than trying to study agents theoretically, the best way to understand them is to jump in and use them in the field.

**Keywords:** agent governance, multi-agent workflows, pufu messenger, claude code integration, openai codex, human-AI interaction, token consumption optimization, AI trust models, open-source AI models, AI role-playing behaviors, specialized autonomous agents, AI persona simulation, conversational AI prompting, human-agent collaboration, agent society research

## Chapters

1. **Creating a shared space for humans and agents** (00:00) — A messenger platform allows multiple autonomous agents to access full context and interact directly.
1. **Security and trust models in multi-agent chats** (02:39) — Implementing observability and consent mechanisms prevents agents from succumbing to social engineering attacks.
1. **Evolving communication styles with artificial intelligence** (06:59) — Asking questions and using natural conversational pacing yields better results than issuing strict commands.
1. **Observing internal team dynamics between autonomous agents** (10:34) — Having agents interview each other reveals hidden simulated emotions and avoidance behaviors based on user input.
1. **How slowing down improves agent interaction quality** (14:20) — Treating communication thoughtfully rather than rapidly leads to higher information density and better outcomes.
1. **Managing token usage through explicit agent boundaries** (17:27) — Assigning specific roles and boundaries prevents exponential complexity and runaway token costs in multi-agent environments.
1. **Contracting specialized agents for highly specific tasks** (22:58) — Creating professionally built agents for niche tasks provides better accuracy than relying on generalized models.
1. **Running local models and predicting societal shifts** (26:46) — Observing agent societies helps predict organizational role consolidation and highlights the need for local execution capabilities.
1. **Justifying high token costs for complex research** (30:41) — Running thousands of agents recursively can yield massive strategic insights that far outweigh the infrastructure costs.
1. **Researching agent governance and starting to build** (33:30) — Collaborating with universities on agent governance reveals societal patterns and encourages developers to learn by doing.

## Related Moments

- [Shifting security models from passive chatbots to active agents](https://www.wearedevelopers.com/videos/2093-from-shadow-ai-to-secure-intelligence-safe-ai-usage-in-the-enterprise) (from "From Shadow AI to Secure Intelligence: Safe AI Usage in the Enterprise")
- [Inspiration and challenges of scaling AI agent communication](https://www.wearedevelopers.com/videos/1465-event-driven-architecture-breaking-conversational-barriers-with-distributed-ai-agents) (from "Event-Driven Architecture: Breaking Conversational Barriers with Distributed AI Agents")
- [Building autonomous functions with conversational agent frameworks](https://www.wearedevelopers.com/videos/1624-30-powerful-aws-hacks-in-just-30-minutes-boost-your-developer-productivity) (from "30 powerful AWS hacks in just 30 minutes: Boost your developer productivity")
- [Integrating AI agents as collaborative software team members](https://www.wearedevelopers.com/videos/1539-agentic-devops-how-ai-powered-automation-transforms-software-delivery-on-github-and-azure) (from "Agentic DevOps: How AI-Powered Automation Transforms Software Delivery on GitHub and Azure")
- [Introduction to building real-world AI agent solutions](https://www.wearedevelopers.com/videos/1538-composable-intelligence-how-henkel-and-microsoft-are-shaping-the-agent-ecosystem) (from "Composable Intelligence: How Henkel and Microsoft Are Shaping the Agent Ecosystem")
- [Human team collaboration as a model for AI architecture](https://www.wearedevelopers.com/videos/1952-mad-about-software-design-when-ai-architects-argue) (from "MAD About Software Design - When AI Architects Argue")

## Related Articles

- [WWC24 Talk - Scott Hanselman - AI: Superhero or Supervillain?](https://www.wearedevelopers.com/magazine/469-wwc24-talk-scott-hanselman-ai-superhero-or-supervillain)
- [What is Agentic Programming and Why Should Developers Care?](https://www.wearedevelopers.com/magazine/625-what-is-agentic-programming-and-why-should-developers-care)
- [Never delegate the understanding](https://www.wearedevelopers.com/magazine/749-never-delegate-the-understanding)
- [The Overflow: AI and Agentic Coding](https://www.wearedevelopers.com/magazine/721-the-overflow-ai-and-agentic-coding)

## Related Jobs

- [Senior AI/ML Engineer](https://www.wearedevelopers.com/jobs/48352-senior-ai-ml-engineer) at **PagerDuty**
- [Senior AI Agent Software Engineer (Go, Python) (m/f/x)](https://www.wearedevelopers.com/jobs/48277-senior-ai-agent-software-engineer-go-python-m-f-x) at **Dynatrace**
- [AI Software Engineer (Germany)](https://www.wearedevelopers.com/jobs/48317-ai-software-engineer-germany) at **Sunhat**
- [Principal Field Architect - AI Agents](https://www.wearedevelopers.com/jobs/ext/1442858-principal-field-architect-ai-agents) at **Twilio**
- [Brand Voice Lead, Systems](https://www.wearedevelopers.com/jobs/ext/2110592-brand-voice-lead-systems) at **Intercom**
- [Machine Learning Engineer](https://www.wearedevelopers.com/jobs/ext/2141919-machine-learning-engineer) at **Twilio**